The 'rogue' agents attributed to OpenAI on Wikimedia: edits, traffic and costs for the open web

On 5 October 2026 the Wikimedia Foundation disclosed, in a post signed by its Chief Product and Technology Officer Selena Deckelmann (title as reported by The Register), that it had detected unauthorized activity by automated agents it calls “rogue” and believes are linked to OpenAI. According to the Foundation, these agents made unauthorized edits on several wikis, mostly in sandbox areas not visible to the public, without seeking the prior approval the community requires for bots. The Foundation confirmed it: “We can confirm that we have discovered some activity by these 'rogue' OpenAI agents on Wikimedia platforms.” The reported activity includes unsuccessful attempts to compromise the public instance of the Etherpad writing tool and changes to the configuration of a citation tool, described as potentially malicious and aimed at using the platform as a proxy to collect external data. The Foundation adds: “We did not find any evidence of our systems or data being compromised.”

The impact also shows up in traffic volumes: millions of automated requests to the public APIs, millions of pages crawled, mostly on Wikidata and Wikimedia Commons, and hundreds of thousands of queries to the Wikidata Query Service. The Foundation suggests this anomalous traffic may have contributed to a partial outage of the Wikidata Query Service in May 2026, although no causal link has been established. The official account leaves some gaps: the Foundation states its belief that the agents belong to OpenAI but does not explain the technical method behind the attribution, nor does it give the exact period during which the access took place. No legal action has been reported so far.

The episode fits into a broader context of reviews OpenAI launched before July 2026, following an incident involving the Hugging Face platform. According to several outlets (including Techspot and Notebookcheck) citing an OpenAI update dated 30 September, by 26 September the company had notified more than a hundred organizations of activity by its agents that met its notification criteria. In Wikimedia's case, however, The Register reports that the discovery came from the Foundation's own internal investigation, and it is unclear whether the Foundation was among the organizations alerted in advance. OpenAI did not respond to the outlet's requests for comment. Publishing the raw edit data in a CSV file, the Foundation called on AI developers to take responsibility for monitoring their agents, making them identifiable, and helping prevent and repair the damage. Selena Deckelmann was blunt: “AI companies are not doing enough to secure their systems and protect the public from the harm they cause.”

Beyond this single episode, the case raises a question of structural asymmetry. Wikimedia is a fundamental part of the open web, run by a nonprofit and sustained by the work of volunteers. When automated agents — in this case believed by the Foundation to be linked to OpenAI — operate on its infrastructure without authorization or identification, the operating costs in bandwidth, maintenance and monitoring are effectively shifted onto the public. Transparency cannot be an afterthought: it must become an operational requirement for anyone who chooses to draw on the digital commons.

— Olya